A bachelor's thesis on author identification found that simple word-usage statistics could reveal who wrote a short paragraph. Readers rely on more than word count: patterns, sentence structure and the way a story is told. These cues make a writer's style easy to feel, and that is what makes a text human.
AI-generated text averages style, so the author behind it becomes hard to see, even with various prompt tricks. Without that style, empathy is harder. When you know a writer personally, you can imagine their perspective; when you don't, non-AI text still reveals something about the person and their context. With AI, that context is ironically lost.
AI writing also loses precision. Even with a detailed prompt, a model must make assumptions to stitch a text together. Human writers face a similar problem when writing for others, often choosing vague language to avoid logical errors. Writing on paper also helps people spot inconsistencies and gaps in their reasoning, an opportunity AI removes by jumping straight to a result.
The loss of style and precision slows readers down. The author feels that AI-generated text takes at least a third longer to read, even at the same length, and most models are extremely verbose, which raises total reading time. Writers may save time, but the cost shifts to readers, along with more questions, more iterations and less trust between readers and the writer.
Source: Hacker News · Summarized by HeadlinesBriefing